Two UROPS projects and a final-year project, all in the Ungur group. Newest first.
Final-year project, CM4288. AY2025/26 Semester 2 to present.
A reactive machine-learning potential for iron oxidation
The aim is to simulate the oxidation of iron with near-DFT accuracy on time and length scales that DFT itself cannot reach. I generate reference data with periodic DFT in Quantum ESPRESSO and use it to train a machine-learning interatomic potential. The potential has to handle bond breaking and the growth of the oxide layer.
Quantum ESPRESSO, machine-learning interatomic potentials, reactive MD.
UROPS, CM3288 and CM3289. AY2024/25 Semester 2 to AY2025/26 Semester 1.
Benzene and ethanol on Cu(100) and Cu(111)
I simulated a benzene/ethanol mixture in contact with copper surfaces using classical MD in LAMMPS. To separate the effect of the force field from the effect of the long-range electrostatics, I ran four combinations (OPLS-AA or TraPPE-UA, each with PPPM or MSM) through the same five-stage protocol: soft-potential relaxation, energy minimisation, staged heating, equilibration and production.
I wrote the Python tooling for setting up and analysing the runs (ASE, NumPy, SciPy, pandas, matplotlib): radial distribution functions, density profiles, surface excess and interfacial tension, averaged over replicate simulations. I presented intermediate results as a poster at the NUS Science Summer Institute 2025 symposium. The Korean LAMMPS guide grew out of this project.
A later check found that the copper slab in the PPPM/MSM comparison run was not a correct fcc(100) surface, so absolute first-layer values from that run should not be read as Cu(100) values. The comparison between the two methods still holds because both used the same slab. Details are in the PPPM and MSM note.
LAMMPS, OPLS-AA, TraPPE-UA, PPPM, MSM, Python.
UROPS, CM2288. AY2023/24 Semester 2.
Exchange and anisotropy in a Dy3 triangle
The molecule is a triangular Dy3 cluster with C3 symmetry, a candidate for a toroidal arrangement of magnetic moments. I calculated its low-lying energy spectrum, the exchange between the Dy ions, and its magnetic response.
A full-cluster multireference calculation was too expensive, so I used two shortcuts. The first replaces two of the three Dy(III) ions with closed-shell Lu(III) and uses the C3 symmetry for the other two sites. The second computes the metal centre and the ligand shell separately and merges their molecular orbitals.
For the Dy site I ran CASSCF(9,7) over the 4f shell (21 spin–orbit configurations) with the scalar-relativistic DKH Hamiltonian and the def2-SVP basis, added NEVPT2 dynamic correlation and second-order spin–orbit coupling via QDPT in ORCA, and extracted single-ion anisotropy tensors with SINGLE_ANISO.
I then modelled the cluster in POLY_ANISO with a Heisenberg exchange Hamiltonian on pseudospins S* = 1/2 per site, scanning the isotropic exchange J from −0.35 to −0.75 cm−1. The best fit is J ≈ −0.50 to −0.55 cm−1 (antiferromagnetic), with Ising-limit single-ion anisotropy (gx, gy ≈ 0). With these parameters the calculated χT(T) from 0 to 300 K and M(H) at 1.9 K match unpublished experimental data, which supports a non-collinear, toroidal arrangement of the local moments in the Dy3 plane.
ORCA (CASSCF, NEVPT2, QDPT), SINGLE_ANISO, POLY_ANISO.